backpropagation
Error signal descent
Backpropagation is the algorithm that trains neural networks by working backward from output to input.
When a network produces the wrong answer, backpropagation calculates how much each connection contributed to the error, then adjusts those connections to reduce future mistakes.
The process repeats across thousands of examples until the network learns: The model improved after backpropagation through millions of training samples.
Product of mass, velocity
- regularizationPenalty preventing model complexity
- hyperparameterPre-set training variable
- tuningAdjusting pitch accuracy
- convergenceProcess of approaching sameness
epochs · annealing · overfitting · underfitting · generalization
Heavy objects for lifting
- biasesUnfair inclinations toward something
- perceptronLinear binary classifier algorithm
- multilayerHaving multiple strata
- neuronsSignal-transmitting brain cells
gradient · delta · sigmoid · node · convolutional
Control anticipating disturbances
- recurrentHappening multiple times
- iterationRepeated procedural pass
- convergenceProcess of approaching sameness
momentum
Product of mass, velocity
- regularizationPenalty preventing model complexity
- hyperparameterPre-set training variable
- tuningAdjusting pitch accuracy
convergence · epochs · annealing · overfitting · underfitting · generalization
weights
Heavy objects for lifting
- biasesUnfair inclinations toward something
- perceptronLinear binary classifier algorithm
- multilayerHaving multiple strata
feedforward
Control anticipating disturbances
- recurrentHappening multiple times
- iterationRepeated procedural pass
- convergenceProcess of approaching sameness
synapse
Nerve signal crossing point
cortex · brain · plasticity · cognitive
propagate, propagation, propagated, propagating, propagates, propagator, propagative, backpropagated